Finding Dutch natives in online forums.

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Machine Learning

Background:

• Determining user nationality online is challenging due to the borderless nature of the internet.
• Law enforcement faces difficulties in establishing jurisdiction for online criminal behavior.
• Manual investigation of user nationality on internet forums is arduous and time-consuming.

Purpose of the Study:

• To develop a machine learning model for detecting Dutch native authors of English text.
• To overcome the challenges of manually identifying user nationality in online contexts.
• To assist in establishing jurisdiction for online law enforcement activities.

Main Methods:

• Collected a large corpus of English forum posts from Dutch and non-Dutch authors on Reddit.
• Utilized a bag-of-words representation to capture linguistic features indicative of native language.
• Compared linear support vector machine and regularized logistic regression models.
• Evaluated model performance using f1 score, precision, and average precision metrics.

Main Results:

• Logistic regression with frequency-based feature selection demonstrated the highest performance in predicting Dutch native authors.
• The developed model effectively identified linguistic patterns characteristic of Dutch native speakers writing in English.
• The machine learning approach proved more efficient than manual investigation.

Conclusions:

• The study successfully developed a machine learning detector for identifying Dutch native English speakers.
• Logistic regression with appropriate feature selection is a promising method for this task.
• Further research is needed to assess the general applicability of these models to other online forums.